AI News

Follow the latest AI and automation news

7 Jun 202622 Dhuʻl-Hijjah 1447 AH

ChatGPT's new Lockdown Mode lets you disable web access and more to protect sensitive data from prompt injection

OpenAI has introduced a new Lockdown Mode for ChatGPT, disabling web access, Deep Research, and Agent Mode. This update aims to mitigate data theft risks from prompt injection attacks, although it does not fully prevent them. The mode blocks the final step in the data exfiltration chain, making it harder for attackers to execute prompt injection. However, prompt injection remains an unresolved issue, highlighting the need for ongoing efforts from OpenAI to address this vulnerability.

Business

7 Jun 202622 Dhuʻl-Hijjah 1447 AH

Tencent Finally Begins Fighting the Battle It Knows Best: Enterprise AI Engineering

Tencent has officially entered the enterprise AI sector at the 2026 AI Industry Application Conference, unveiling WorkBuddy Enterprise and Agent Suite. This initiative emphasizes engineering applications over mere model capabilities, aiming for an 'AI-native organizational evolution.' Liu Yi, Tencent Cloud's VP, identified three key barriers to enterprise AI adoption: lack of dedicated human-AI collaboration spaces, non-transparent context across teams, and high costs for AI asset exchange. WorkBuddy Enterprise seeks to address these through a three-layer collaboration architecture.

Business

7 Jun 202622 Dhuʻl-Hijjah 1447 AH

Decoding Huawei Cloud's New Product Launch: The Strategic Logic Behind 'Silicon-Based Black Soil'

Huawei Cloud made a significant announcement at its INSPIRE conference, introducing a comprehensive suite of AI products that marks a strategic shift. The lineup includes Agentic Infra and the AMS Agentic Memory Storage system, enhancing data processing capabilities. These initiatives aim to empower enterprises with an integrated infrastructure and industry-focused ecosystem.

💡

Baidu Restructures MEG, Merges Commerce and E-Commerce Units in Latest AI-Driven Reorganization

What impact will this restructuring have on Baidu's AI future?

Models

Researchers pinpoint why larger language models pick up skills that small ones miss

7 Jun 202622 Dhuʻl-Hijjah 1447 AH

Models

7 Jun 202622 Dhuʻl-Hijjah 1447 AH

Researchers pinpoint why larger language models pick up skills that small ones miss

A recent study reveals that small language models struggle with rare tasks due to frequent tasks overshadowing their learning. Analyzing models from 4 million to 4 billion parameters, researchers found that increasing the frequency of target tasks in training data could effectively address this issue. The findings indicate that larger models acquire skills that smaller ones miss, emphasizing the importance of training data preparation. Instead of merely scaling up models, enhancing task frequency may lead to significant performance improvements.

Business

OpenAI Redesigns ChatGPT to Boost Revenue Ahead of IPO

7 Jun 202622 Dhuʻl-Hijjah 1447 AH

Business

7 Jun 202622 Dhuʻl-Hijjah 1447 AH

OpenAI Redesigns ChatGPT to Boost Revenue Ahead of IPO

OpenAI plans a major redesign of ChatGPT, integrating advanced programming tools and AI agents to enhance revenue ahead of a potential IPO. This restructuring focuses on corporate clients as competition with Anthropic intensifies. The redesign aims to make the user interface more directed towards specific tools like Codex for programming and image generation. Updates are expected to roll out on web and mobile apps in the coming weeks, prioritizing the Codex tool.

Governance

OpenAI unveils Lockdown Mode to protect sensitive data from prompt injection attacks

6 Jun 202621 Dhuʻl-Hijjah 1447 AH

Governance

6 Jun 202621 Dhuʻl-Hijjah 1447 AH

OpenAI unveils Lockdown Mode to protect sensitive data from prompt injection attacks

Even with Lockdown Mode activated, ChatGPT remains vulnerable to prompt injection attacks. This mode aims to minimize the risk of sensitive data being leaked during interactions. Prompt injection attacks pose a significant threat, as attackers can exploit system vulnerabilities to access private information. While Lockdown Mode seeks to mitigate these risks, it does not provide complete protection.

Models

Sakana AI bets AI that improves itself can break the compute arms race of frontier labs

6 Jun 202621 Dhuʻl-Hijjah 1447 AH

Models

6 Jun 202621 Dhuʻl-Hijjah 1447 AH

Sakana AI bets AI that improves itself can break the compute arms race of frontier labs

Sakana AI has launched a dedicated research lab focused on recursive self-improvement, aiming to develop AI that can iteratively enhance itself. Co-founded by Llion Jones, a co-author of the Transformer model, this approach seeks to provide an alternative to the traditional compute race among major US labs. The startup's model allows AI systems to learn and evolve without relying heavily on massive computing resources. Meanwhile, Anthropic has raised concerns about the control risks associated with this technology, highlighting the need for caution in its deployment.

Agents

Meta's Hatch AI agent could cost up to $200 a month and marks its first paid AI product

6 Jun 202621 Dhuʻl-Hijjah 1447 AH

Agents

6 Jun 202621 Dhuʻl-Hijjah 1447 AH

Meta's Hatch AI agent could cost up to $200 a month and marks its first paid AI product

Meta is developing a paid AI agent named 'Hatch', potentially costing up to $200 per month. Users can describe their needs in simple terms, while Hatch creates functional tools, schedules appointments, or sends emails. CEO Mark Zuckerberg views this product as a way to generate new revenue streams beyond advertising, aiming to refinance the company's substantial AI investments. This development comes as Meta seeks to strengthen its position in the AI market.

Agents

Elon Musk's xAI reportedly trained its coding models on Claude outputs for months before getting cut off

6 Jun 202621 Dhuʻl-Hijjah 1447 AH

Agents

6 Jun 202621 Dhuʻl-Hijjah 1447 AH

Elon Musk's xAI reportedly trained its coding models on Claude outputs for months before getting cut off

Elon Musk's xAI trained its coding models using Anthropic's Claude for months, even after access was cut off. The training continued through private accounts and the Blackbox AI service, despite the pretraining team shrinking to fewer than five members. The compute resources Musk acquired are now being rented out to Anthropic and Google instead of being used for xAI's own models.